Mykhailo Volkov

mvolkov-skills

SQLAlchemy 2.0 async-first best practices — engine and pool config, Mapped[T] models, relationship loading strategies (selectin/select/write_only), query optimization, async sessions and transactions, FastAPI dependency injection

FastAPI production best practices — domain-organized project structure, async correctness (no blocking I/O, no CPU work in BackgroundTasks), DI patterns with per-request caching and chaining, Pydantic v2 split create/update/response models, Pydantic Settings v2 with .env and lru_cache, async testing with httpx.AsyncClient

pytest async-first testing best practices — test layer structure (unit/integration/e2e), fixture design with proper scope, async testing with pytest-asyncio (asyncio_mode=auto), test data with factory_boy + faker, HTTP mocking with respx, time mocking with freezegun, property-based with hypothesis, DB transaction-rollback isolation, flush() pattern for chained DB fixtures, pytest-xdist parallelism

Python gRPC async-first Kubernetes-native best practices using grpc.aio — server bootstrap with graceful shutdown that flips Health to NOT_SERVING before server.stop(), shared per-process channel factory, canonical decorator pattern (@grpc_logger + @grpc_error_handler) for cross-cutting concerns, max_connection_age_* triad for HPA rebalancing, client-side LB via dns:/// + round_robin + service config JSON over a Kubernetes headless Service (no service mesh), ERROR_MAP + google.rpc.Status rich error model, client-side translation hierarchy

Git workflow best practices — Git Flow branch prefixes (feature/bugfix/hotfix/release), Conventional Commits format with feat/fix/docs/style/refactor/test/chore types, atomic commits with imperative mood, pre-commit hygiene (tests/diff/lint/no-debug/no-secrets), pull --rebase before push, no force-push to shared branches, max ~400 LOC per PR, self-review first, squash on merge for clean main history

Python observability best practices using OpenTelemetry — SDK bootstrap with off-switch contract and idempotency guard, auto-instrumentation (gRPC/SQLAlchemy/FastAPI/Logging), the LoggingInstrumentor + structlog correlation pattern with custom log hook and CleanLoggingHandler, two-tier log field taxonomy, span attribute conventions with <service>.<key> prefix for cross-service Tempo search, sensitive-field redaction at the serialization boundary, status taxonomy, metrics cardinality control, tail-based sampling, and Sentry-with-OTel integration

Engineering best practices for money / payments / ledger systems — Decimal or integer minor units (never float), the two-layer idempotency model (client-driven idempotency_key + chain-CAS for state transitions, never conflated), double-entry ledger architecture (Accounts + Transfers, debit=credit invariant at DB level, append-only, balance computed not stored), two-phase transfers via HOLD balance, atomic chains for composite operations, transaction state machines with optimistic concurrency control, external payment provider integration via stateless proxy + sync request + async webhook with three-layer dedup, reversibility via separate transaction with parent_transaction_id, and DB-enforced invariants